Uncovering Clinically Relevant Gene Fusions with Integrated Genomic and Transcriptomic Profiling of Metastatic Cancers

With the advent of improved genomic sequencing technologies, structural rearrangements in cancer genomes have been detected and targeted with treatments (1). Since the initial discovery of the Philadelphia chromosome and the BCR-ABL fusion in chronic myeloid leukemia, gene fusion events have been identified in many solid tumors, and serve as effective therapeutic targets (2–5). Some, such as those involving ALK, are specific for a particular cancer type, while others such as NRG1 are observed pan-cancer. Fusions have greatly expanded our knowledge about oncogenesis and provided opportunities for development of treatments (6).

Commercially available panels are now used routinely to help guide clinical care and treatment decision making. These assays can identify preselected fusions of interest, based on our current understanding of pathogenic and clinically relevant fusion events. This type of guided approach is restricted by currently available knowledge and has limitations in fusion discovery. In contrast, WGS and RNA-seq applies next-generation sequencing to offer an unbiased, genome-wide approach to revealing structural rearrangements. This may uncover known fusion events in unexpected tumor histologies, and also identify novel fusion events with potential clinical relevance.

The Personalized OncoGenomics (POG) program at BC Cancer is a research study that integrates whole-genome (WGS) and transcriptome sequencing (RNA-seq) with bioinformatic analysis and a molecular tumor board to identify genomic information in metastatic or recurrent cancers with the goal of understanding the genomic and transcriptomic landscape of metastatic and pretreated cancers, and potentially identifying cancer vulnerabilities amenable to treatment (7, 8). In this study, our objectives were to summarize the findings of fusion detection in the first 570 fully sequenced cases of the POG program, and to compare the fusion prevalence detected by WGS and RNA-seq with that in silico predicted from commercially available panels. We aimed to delineate the individual roles of WGS and RNA-seq, as well as the value of integrative analysis in deciphering clinically relevant events and confirming results. We also correlated the incidence of fusion events with pathologic diagnoses to provide a landscape of fusion events across histologies.

In the Canadian province of British Columbia, the POG program is a translational research study composed of a multidisciplinary network of oncologists, pathologists, genome scientists, and bioinformaticians applying WGS and RNA-seq analysis to guide treatment decision making for patients with advanced or metastatic malignancies. The POG program is registered under clinical trial number NCT02155621 and was approved by the University of British Columbia (Vancouver, Canada)—BC Cancer Research Ethics Board (H12-00137, H14-00681), and conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from patients. A total of 570 patients enrolled in the POG program between July 1, 2012 and August 17, 2017 who had a fresh tumor biopsy and underwent DNA and RNA sequencing were included in this study (8). Baseline clinicopathologic data and treatment outcomes were extracted by retrospective chart review.

Fresh tumor biopsies and blood samples underwent WGS (median depth 84× tumor; 42× matched normal) and RNA-seq (median 203 million reads) followed by in-depth bioinformatic analysis (8). This integrative use of WGS and RNA-seq, in conjunction with bespoke bioinformatics tools and analytic pipelines, informs on the molecular pathogenesis of tumors and also predicts potential therapeutic targets that are reviewed at a molecular tumor board.

Structural variants (SV) in RNA-seq data were identified using ABySS (v1.3.4), TransABySS (v1.4.10), Chimerascan (v0.4.5), and DeFuse (v0.6.2; refs. 9–13). SVs in WGS were identified using ABySS, Trans ABySS, Manta (v1.0.0), and Delly (v0.7.3; refs. 14, 15). Putative SV calls were aggregated and computationally validated using consensus caller MAVIS for Merging, Annotation, Validation, and Illustration of Structural variants (v2.1.1), and then additionally filtered to identify those called by more than one tool, and for which a breakpoint contig could be assembled (16). DNA events with identical genomic breakpoints in more than one sample were also excluded as potentially confounding germline variants and technical artifacts. SVs were annotated based on Ensembl gene models (v69; ref. 17). Fusion products of interest were manually reviewed using the Integrative Genomics Viewer to evaluate read evidence and presence of genomic features such as repeats (18). Select recurrent fusions of interest were orthogonally validated using quantitative polymerase chain reaction (PCR) and custom probes (Supplemental Data).

RNA-seq reads were processed using Jaguar (v2.0.3; ref. 19). Expression by gene was computed in reads per kilobase per million mapped reads (RPKM) based on Ensembl gene models (v69; ref. 17). RPKMs of NTRK genes were compared with RPKMs computed from The Cancer Genome Atlas (TCGA) gene expression data (level 3; https://tcga-data.nci.nih.gov/tcga/; ref. 20). Each gene in the patient tumor sample was compared with a ranked list of all TCGA expression values for that gene to assign it an overall tumor percentile.

We focused on recurrent fusions, which were defined as fusions made up of the same gene pairs, that were detected in at least two patients. Recurrent fusions were first identified from the entire cohort of 570 patients, and then prevalence of recurrent fusions by each tumor site was calculated. We also compared the recurrent fusion prevalence detected by WGS and RNA-seq with those predicted to be detected in silico from commercially available panels. Running all four commercially available panels on every case was not practical in terms of available patient samples. Specifically, we examined commercially available panels that listed detected fusions on their respective websites, including: Archer Solid Tumor FusionPlex, FoundationOne CDx, Oncomine Comprehensive Assay version 3M, and the Guardant360 Liquid Biopsy Assay (73-gene version; see Supplementary Table S1). Although these panels do not cover all potential breakpoint partners of each gene, we presumed if a gene was listed in their available documentation that any potential fusion that contained that gene could be detected by the panel. This was also necessary as many commercially available assays do not provide full annotation of what fusion partners and sites within a given gene are detectable. To confirm whether a fusion partner was novel, we compared the identified fusion partners with available literature and a database of tumor fusions identified in TCGA (21).

Homologous recombination deficiency (HRD) score was defined by the sum of loss of heterozygosity, telomeric allelic imbalance, and large-scale state transitions measuring ≥34 as reported previously (22–24). Microsatellite instability (MSI) was defined as an MSI-sensor score ≥30 (25).

Descriptive statistics characterized patients who underwent WGS and RNA-seq within the POG program. Pearson χ² or Fisher exact test were conducted to determine whether there was an association between fusion prevalence and patient characteristics as appropriate. All tests were two sided, with P ≤ 0.05 as the cutoff for statistical significance. Stata version 15.1 was used for all statistical analyses.

Of 570 patients enrolled in the POG study, 25 histologies were included, with biopsies taken from 18 organ groups (8). 544/570 (95%) of patients were found to have at least one fusion by WGS and 172/570 patients (30%) had recurrent fusion events that were detected by WGS in at least two patients (Fig. 1). We focused on recurrent fusion events as they are often indicative of oncogenic drivers (26). MAVIS identified 81 unique fusions by WGS and 111 by RNA-seq that were recurrent within the cohort, of which 18 fusions by WGS and 19 by RNA-seq were noted in at least three separate tumors. There were eight overlapping fusions that were recurrent in both DNA and RNA (EWSR1-PATZ1, FLI1-EWSR1, GAK-TACC3, NACC1-NIPBL, PDLIM5-BMPR1B, PLB1-PP1CB, TACC3-FGFR3, UBR2-PTK7) and three fusions that overlapped between DNA and RNA and were noted in at least three samples (MIPOL1-TTC6, EML4-ALK, MYB-NFIB). A selection of recurrent but not well-described fusions, including SLC4A2-AGAP3 and TACC3-FGFR3, were validated with an orthogonal custom PCR assay and shown to be present in all tested tumors (Supplemental Data). Multiple previously reported NRG1 fusions with different fusion partners were noted but not included in this list to avoid bias despite being clinically actionable, as the different fusion partners thus did not meet the definition of recurrence used in this analysis (27, 28). Only an ATP1B1-NRG1 fusion was found to be recurrent for fusions involving NRG1. The most common fusions were EML4-ALK in thoracic malignancies (9/69, 13% prevalence), and CMTM8-CMTM7 in colorectal cancer (4/73, 5.5% prevalence). Fusion prevalence by tumor site is further detailed in Table 1. Fusion prevalence was not significantly associated with clinical characteristics, including gender (P = 0.10), age at diagnosis (P = 0.76), or tumor type (P = 0.44). There was no significant association between fusion prevalence and either HRD or MSI status (P = 0.35, P = 0.27, respectively).

Several recurrent fusions were only detected in RNA and not in DNA (Fig. 1), including CMTM8-CMTM7 and RRM2-C2orf48 in colorectal cancer. RRM2-C2orf48 was highly recurrent, seen in 5 patients with colorectal cancer, 1 with pancreatic ductal adenocarcinoma, and 1 with anal squamous cell carcinoma. Detection of fusions in RNA only may relate to complex genomic events or sequence alignment considerations making detection in DNA difficult, but there is evidence that a subset of these events only exist in the RNA within the tumor due to phenomena such as trans-splicing between gene transcripts (29).

We also manually reviewed fusions identified involving tyrosine kinase domains (TKD): CD74-ROS1, TACC3-FGFR3, EML4-ALK, and PRKAA1-TTC33. All were predicted to be in-frame and contain a TKD; thus, these plausibly represent oncogenic fusions.

Differences in fusion prevalence detected by MAVIS compared to in-silico predicted prevalence among commercial panels are visualized in Fig. 2. In comparison to MAVIS, the Archer panel was predicted to detect only 8/111 fusions (7.2%; P < 0.01), specifically those involving ALK, ESR1, EWSR1, FGFR2, FGFR3, MYB, and NUMBL as a fusion partner. The FoundationOne panel was predicted to detect 6/111 fusions (5.4%, P < 0.01) involving ALK, BCR, EWSR1, FGFR2, FGFR3, and MYB. The Thermo Fisher Oncomine Comprehensive Assay v3 was predicted to detect 6/111 (5.4%, P < 0.01), involving ALK, ESR1, FGFR2, FGFR3, MYB, and PTEN. With the Guardant360 assay, 3/111 fusions (2.7%; P < 0.01) are predicted to be detected, involving ALK, FGFR2, and FGFR3. In addition to differing coverage for fusion gene partners, many of the panels have different sensitivity and specificity for fusion calls, so it is difficult to compare these assays directly. We have also made a broad assumption that any gene listed within the panel would be detected for the purposes of this in silico predicted prevalence comparison.

The WGS and RNA-sSeq approach was able to detect potentially clinically actionable fusions that are unlikely to be included on commercial assays. In addition to the well-characterized TPM3-NTRK1 fusion, we identified novel fusion partners for NTRK1 (SLC25A44), NTRK2 (SHC3, DAPK1), and NTRK3 (POLG, PIBF1; Fig. 3). These novel fusion partners have not been previously reported, including in a recent publication of the largest cohort of TRK fusion–positive cancers or in TCGA (tumorfusions.org; refs. 21, 30). The percentile of NTRK expression is shown in comparison with the average expression across all histologies in the TCGA. Expression diagrams are also shown to illustrate the expression level across the entire predicted fusion transcript. The TPM3-NTRK1 fusion identified in a patient with breast cancer in this cohort displayed a transcript structure and expression pattern characteristic of an oncogenic fusion, with a dramatic increase in expression in the 3′ end of NTRK1 downstream of the breakpoint. Three of five novel NTRK fusions detected are predicted to include the TKD as part of the fusion product and display above average expression, suggestive of oncogenic fusions. Conversely, the NTRK3-POLG and SLC25A44-NTRK1 fusions are not predicted to include the TKD and have low expression. These results exemplify the utility of deep RNA sequencing for the detection and interpretation of novel fusion events. None of these 6 patients were treated with an NTRK inhibitor as these agents were not available at the time of analysis, with the earliest case identified in early 2014.

In addition to enhancing the discovery process for potentially pathogenic fusions, WGS and RNA-seq can also identify cryptic or unexpected variants of well-established fusions. In one case in our study, a patient with metastatic non–small cell lung cancer harbored an EML4-ALK fusion that was identified only by WGS and not detected by parallel ALK IHC and FISH analysis (Fig. 4). ALK IHC was equivocal +2/3 and break-apart FISH was negative with 3/100 positive cells only. With the initial negative ALK testing, he was treated with carboplatin and gemcitabine, a clinical trial of pemetrexed and reolysin, and erlotinib, all with short-interval disease progression. WGS subsequently identified a well-described inversion event in chromosome 2, resulting in the EML4-ALK fusion gene variant 1, but a second translocation event was also identified involving the insertion of a 55 Mb portion of chromosome 2p containing the EML4-ALK fusion (2p23.2–2p11.2) into chromosome 12 (12q12), which likely contributed to the negative FISH result. This event was further supported by copy-number analyses showing copy-number gains corresponding to the region of the 55Mb sequence translocated from chromosome 2 to 12 (Fig. 5).

This patient received crizotinib with excellent response, but then developed radiographic progression at 7 months. Repeat biopsy of a liver metastasis demonstrated gain of ALK copy number (4 copies as compared with 3 copies in the pre-crizotinib tumor). Second-line ALK inhibitor therapy with ceritinib was continued for another 7 months before treatment discontinuation for clinical and radiographic progression. At the time of treatment, third-generation ALK inhibitors were not yet available, but the use of ALK inhibitor therapy prolonged survival by an additional 14 months.

Molecular profiling has altered the landscape of oncology, with gene fusions revolutionizing treatments in particular tumor types. WGS and RNA-seq provides a wealth of information to better delineate the genomic landscape of cancers. We report our institutional experience with detection of recurrent fusions, demonstrating that WGS and RNA-seq can detect both known and novel fusions, only 10% of which are predicted to be detectable by current commercially available panels. The discovery of NRG1 fusions previously reported by the POG investigators in pancreatic cancer, as well as other tumor types including lung adenocarcinoma and cholangiocarcinoma, highlights a practice-changing discovery yielded by WGS and RNA-seq technologies (27, 28). In this article, our detection of novel fusion partners for NTRK1, NTRK2, and NTRK3, and a cryptic ALK fusion further features the potential of this approach for clinically relevant fusion discovery.

Within the POG study, the goal of WGS and RNA-seq is a discovery process, in addition to providing clinically relevant and potentially actionable details to patients and their treating clinicians. The discovery component is crucial in identifying events which may be, or become, clinically targetable, as this broadens the scope of patients who can benefit from fusion-targeting therapy. When this leads to discovery of novel treatment approaches, such as with NRG1 fusions, it is possible to query the sequence data already available to retrospectively identify other patients who could benefit from the same therapy, as well as to detect the event in all future patients (27, 28). This obviates the need for a repeat assay, which often requires an additional tissue biopsy. Panel approaches require modification to detect new fusions, which takes development and time, and cannot provide the possibility of retrospective detection without requiring additional testing. This remains a similar barrier with other assays, such as anchored multiplex polymerase chain reaction. Requerying data also allowed us to identify several novel NTRK fusion partners, which were discovered in patients before NTRK testing and therapies were available.

This type of discovery platform incorporating RNA-seq is more robust than genomic data alone, as expressed variants can be identified with two approaches, providing increased sensitivity and confidence in sequencing results. Coupling this type of platform with access to novel therapies that allow n of 1 trials to assess clinical significance, or with a strong functional genomics platform, can provide evidence for novel and emerging therapeutically relevant variants. Many fusion events are rare, so it is unlikely that it would be feasible to conduct a trial for every individual combination identified.

In another scenario, WGS and RNA-seq was able to uncover a novel complex rearrangement of the EML4-ALK fusion oncogene from chromosome 2 to 12. ALK IHC and FISH, both considered standards of care, had previously been negative for this patient. There have been previous reports of similar “atypical” false-negative patterns, highlighting the limitations of FISH methods in these cases (31, 32). As a result, “atypical” negative findings may warrant further detailed genomic evaluation to avoid false-negative results. Broad fusion detection also has the advantage of detecting fusions in unexpected tumor types, and can result in changes in diagnosis, particularly in fusion-driven malignancies like sarcomas (33).

Although whole genome and transcriptome sequencing offers greater breadth in fusion detection, possibilities for discovery, and rapid use of new knowledge, there are several challenges to its use in the clinic. The advantages and disadvantages between a panel-based approach compared to whole genome and transcriptome sequencing have been well described (34–37). Challenges include interpretation of variants of unknown significance, more complex bioinformatics analysis, and considerations of resource availability and economic implications. In patients with metastatic non–small cell lung cancer, a decision analytic model demonstrated cost savings and faster turnaround times with upfront next-generation sequencing compared with sequential single-gene testing (38). As the cost of sequencing decreases, similar economic analyses, particularly with respect to gene fusions, will be helpful in delineating the role of next-generation sequencing in routine clinical practice.

In this analysis, we did not demonstrate a high prevalence of HRD or MSI in patients with recurrent fusions. This may be related to the heterogeneity of our cohort with a series of diverse tumor types, and to the influence of patient selection for inclusion in the POG program. Patients without identifiable drivers on standard testing with adequate performance status were prioritized for inclusion in POG. In colorectal cancer, microsatellite instability-high (MSI-H) status has been associated with increased gene fusions (39–42). Given that our cohort included only 73 cases of colorectal cancer, this sample size likely limits the detection of a similar correlation between MSI-H status and gene fusions and may also explain why we did not see differences based on histology as others have described previously. (30).

In this analysis, to avoid introducing bias, we did not specifically count all NRG1 fusions that were previously reported by POG investigators as recurrent events, given that the fusion partners were different. However, another advantage of WGS and RNA-seq is that even different fusion partners can be taken within a pathway context, so a new fusion partner in an analogous location or with upregulation of similar pathways to a known partner can be trialed therapeutically (27, 28).

Limitations of this study include the heterogeneity of the cohort, encompassing a large number of tumor types to provide a descriptive landscape of fusion detection in our POG program; as a result, estimates of fusion prevalence are based on small numbers of cases in each tumor type. Patients enrolled in this program had an adequate performance status to be eligible for later line therapies, and this may introduce a selection bias. As individuals were enrolled and underwent WGS and RNA-seq at different time points during their disease trajectory, it is possible that the detected fusions developed later in tumor progression or as a resistance mechanism. Panel sequencing was not run concurrently, and so a false-negative rate for fusion detection by WGS and RNA-seq cannot be assessed. Subclonal alterations may not be detectable at the sequencing depth used, and regional tumor heterogeneity is not captured, though most actionable driver events would be expected to be truncal and widespread in the tumor. It is also notable that every fusion clinically detected with standard of care testing prior to WGS and RNA-seq was also observed in the genomic data. We focused on recurrent fusions, as these are considered more likely to have an oncogenic role and thus be selected for in multiple tumors. Functional annotation of all variants is beyond the scope of this article, but future functional validation, particularly of novel NTRK partners and kinase domain containing fusions, would further delineate the role of these fusions in tumor development and progression.

Utilizing WGS/RNA-seq facilitates identification of both known targetable fusions as well as novel fusions in clinically relevant genes, some of which have not yet been characterized. A significant benefit of having WGS and RNA-seq data available for patients is the innate ability to retrospectively identify a variant that becomes clinically relevant over time, whereas technologies that require a priori identification of variants are less flexible in an evolving cancer therapeutic landscape. As sequencing and analysis costs are reduced, our findings lend support for the increased applicability of WGS and RNA-seq in the clinic.

C.J. Grisdale reports grants from NIH during the conduct of the study. R.A. Moore reports grants from BC Cancer Foundation and TFRI during the conduct of the study. H. Lim reports other from Roche, Ipsen, BMS, Merck, Eisai, and Taiho outside the submitted work. D.F. Schaeffer reports personal fees from Alimentiv Inc., Pfizer, Amgen, Diaceutics, Merck and other from Satisfai Health Inc outside the submitted work. D.J. Renouf reports personal fees from Roche, Bayer, Celgene, Servier, Ipsen, Taiho, and AstraZeneca outside the submitted work. S. Yip reports personal fees from Amgen, AstraZeneca, Bayer, Norvatis, Roche, and Pfizer outside the submitted work. J. Laskin reports other from BC Cancer Foundation during the conduct of the study; grants from Roche Canada, and personal fees from Roche Canada, Pfizer, Eli Lilly, and AstraZeneca outside the submitted work. S.J.M. Jones reports grants from BC Cancer Foundation, Terry Fox Research Institute, Canada Foundation for Innovation, Canada Research Chairs, and BC Knowledge Development Fund during the conduct of the study. J.M. Loree reports grants from Michael Smith Health Professional Investigator Award and BC Cancer Foundation during the conduct of the study; grants and personal fees from Ipsen; personal fees from Taiho, Amgen, Novartis, Eisai, Bayer, and Pfizer outside the submitted work. No disclosures were reported by the other authors.

E.S. Tsang: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing-original draft, writing-review and editing. C.J. Grisdale: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. E. Pleasance: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. J.T. Topham: Investigation, visualization, writing-review and editing. K. Mungall: Data curation, investigation, methodology, writing-review and editing. C. Reisle: Data curation, investigation, methodology, writing-review and editing. C. Choo: Data curation, investigation, writing-review and editing. M. Carreira: Data curation, investigation, writing-review and editing. R. Bowlby: Data curation, investigation, writing-review and editing. J.M. Karasinska: Investigation, writing-review and editing. D. MacMillan: Data curation, investigation, writing-review and editing. L.M. Williamson: Data curation, investigation, writing-review and editing. E. Chuah: Data curation, writing-review and editing. R.A. Moore: Data curation, investigation, writing-review and editing. A.J. Mungall: Data curation, investigation, writing-review and editing. Y. Zhao: Data curation, investigation, writing-review and editing. B. Tessier-Cloutier: Data curation, investigation, writing-review and editing. T. Ng: Investigation, writing-review and editing. S. Sun: Investigation, writing-review and editing. H.J. Lim: Investigation, writing-review and editing. D.F. Schaeffer: Resources, investigation, writing-review and editing. D.J. Renouf: Resources, investigation, writing-review and editing. S. Yip: Conceptualization, resources, investigation, writing-original draft, writing-review and editing. J. Laskin: Resources, investigation, writing-review and editing. M.A. Marra: Resources, investigation, writing-review and editing. S.J.M. Jones: Resources, investigation, writing-review and editing. J.M. Loree: Conceptualization, resources, data curation, formal analysis, supervision, investigation, visualization, methodology, writing-original draft, writing-review and editing.

J.M. Loree and D.J. Renouf were recipients of Michael Smith Health Professional Investigator Awards that helped support this work. D.F. Schaeffer is a recipient of the VCHRI Investigator Award. M.A. Marra acknowledges infrastructure investments from the Canada Foundation for Innovation and the support of the Canada Research Chairs and the CIHR Foundation (FDN-143288) programs.

We gratefully acknowledge the participation of the patients enrolled in the POG program and the generous support of the BC Cancer Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.